基于Jaya-BP神经网络的混凝土坝参数反演

Inversion Analysis for Parameters of Concrete Dam Based on Hybrid Jaya-BP Neural Network

针对混凝土坝等大体积混凝土结构性能参数高效反馈问题,提出一种基于Jaya-BP神经网络的新型参数反演算法。采用Jaya算法优化BP神经网络的权值和阈值,克服其迭代收敛速度偏慢、易陷入局部最优解等缺点,提高BP神经网络的全局寻优能力和稳定性。引入正交设计法以及拉丁超立方法设计参数组合,通过有限元分析得到较为准确的训练样本,形成基于Java-BP神经网络结合有限元分析的大体积混凝土结构参数高效反演算法。以典型混凝土重力坝和拱坝为例,对坝体和坝基弹性模量进行反演分析,并与传统的BP神经网络反演结果进行对比表明,Jaya-BP神经网络反演精度明显提高。

A new parameter inverse algorithm based on Jaya-BP neural network is proposed for the inverse analysis of concrete dam with higher accuracy and efficiency.The Jaya algorithm is employed to optimize the weights and thresholds of the classical BP neural network for overcoming its shortcomings in slow convergence speed and falling to locally optimized solutions,and getting higher global optimization ability and stability.Combining with orthogonal design method with Latin hypercube sampling for parameter set,the finite element analysis is employed to obtain more accurate training samples,and an efficient parameter inversion algorithm of mass concrete structure based on Java-BP neural network combined with finite element analysis is formed.Taking typical concrete gravity dam and arch dam as examples,the elastic modulus of dam body and foundation are analyzed by inversion,and the comparison with the traditional BP neural network inversion results shows that the Jaya-BP neural network inversion accuracy is obviously improved.